Re: subexpressions

--- Tijs <tijs_news@bluescraper.nl> wrote:

> Steve Howell wrote:
> > FWIW there's the possibility that even without a
> > subexpression syntax, some Python implementations
> > would detect the duplication of x*x and optimize
> that
> > for you. It would have to know that x*x had no
> side
> > effects, which I think is a safe assumption even
> in a
> > dynamic language like Python.
>
> No, x may be an object that has the __mul__ special
> method, and it may have
> side effects.
>

Ok, I stand corrected.

Duplicate subexpressions are pretty easy to avoid in
Python, so though an optimization would not be
impossible here (checking for immutability of
builtins, etc., which still assumes the idea that
multiplication is more expensive than checking for
immutability even for the common builtin case), it
would not be worthwhile.

Shortly after I posted, there was an elegant solution
to avoiding having to repeat x*x in the lambda, so the
point's kind of moot now.

____________________________________________________________________________________
Shape Yahoo! in your own image. Join our Network Research Panel today! http://surveylink.yahoo.com/gmrs/yah...invite.asp?a=7

> Ok, I stand corrected.
>
> Duplicate subexpressions are pretty easy to avoid in
> Python, so though an optimization would not be
> impossible here (checking for immutability of
> builtins, etc., which still assumes the idea that
> multiplication is more expensive than checking for
> immutability even for the common builtin case), it
> would not be worthwhile.
>
> Shortly after I posted, there was an elegant solution
> to avoiding having to repeat x*x in the lambda, so the
> point's kind of moot now.

The elegance of that solution very much depends on the cost of the duplicate
operation vs. the additional function call.

And for the usecase at hand, that's exactly the point not to do it:

droggisch@ganesha:/tmp$ python -m timeit '(lambda x: lambda y: y+y)(10 *
10)'
1000000 loops, best of 3: 1.04 usec per loop
droggisch@ganesha:/tmp$ python -m timeit 'lambda: 10 * 10 + 10 * 10'
1000000 loops, best of 3: 0.336 usec per loop

Diez

Re: subexpressions - Python